library_name, base_model, tags, datasets, model-index

library_name	base_model	tags	datasets	model-index
transformers	W-61/llama-3-8b-base-sft-hh-harmless-8xh200	alignment-handbook margin-dpo generated_from_trainer	Anthropic/hh-rlhf	name results llama-3-8b-base-margin-dpo-hh-harmless-8xh200-20260410-180850

llama-3-8b-base-margin-dpo-hh-harmless-8xh200-20260410-180850

This model is a fine-tuned version of W-61/llama-3-8b-base-sft-hh-harmless-8xh200 on the Anthropic/hh-rlhf dataset. It achieves the following results on the evaluation set:

• Loss: 0.5388
• Margin Dpo/margin Mean: 7.1205
• Margin Dpo/margin Std: 10.4987
• Logps/chosen: -80.9964
• Logps/rejected: -92.9393
• Logps/ref Chosen: -71.4909
• Logps/ref Rejected: -76.3133
• Logits/chosen: -0.4986
• Logits/rejected: -0.4860

Model description

More information needed

Intended uses & limitations

More information needed

Training and evaluation data

More information needed

Training procedure

Training hyperparameters

The following hyperparameters were used during training:

• learning_rate: 5e-07
• train_batch_size: 16
• eval_batch_size: 16
• seed: 42
• distributed_type: multi-GPU
• num_devices: 8
• total_train_batch_size: 128
• total_eval_batch_size: 128
• optimizer: Use OptimizerNames.ADAMW_TORCH with betas=(0.9,0.999) and epsilon=1e-08 and optimizer_args=No additional optimizer arguments
• lr_scheduler_type: cosine
• lr_scheduler_warmup_ratio: 0.1
• num_epochs: 1

Training results

Training Loss	Epoch	Step	Validation Loss	Margin Dpo/margin Mean	Margin Dpo/margin Std	Logps/chosen	Logps/rejected	Logps/ref Chosen	Logps/ref Rejected	Logits/chosen	Logits/rejected
0.6266	0.3030	100	0.6174	2.2836	3.9974	-75.6156	-82.7216	-71.4909	-76.3133	-0.5741	-0.5577
0.5253	0.6061	200	0.5437	6.4618	9.5445	-79.5821	-90.8664	-71.4909	-76.3133	-0.5200	-0.5068
0.5534	0.9091	300	0.5388	7.1205	10.4987	-80.9964	-92.9393	-71.4909	-76.3133	-0.4986	-0.4860

Framework versions

• Transformers 4.51.0
• Pytorch 2.3.1+cu121
• Datasets 2.21.0
• Tokenizers 0.21.4